Fix wires object handling in Sum.terms method

[andrijapau]

Context:

Prior to this PR, the wires information for Identity was not being retained correctly,

import pennylane as qml

hamiltonian = qml.dot(
        [1, -0.5, 0.5], [qml.Identity(wires=[0, 1]), qml.PauliZ(0), qml.PauliZ(0)]
)
coeffs, ops = hamiltonian.terms()
I, Z = ops

>>> print(coeffs, ops)
[1.0, np.float64(0.0)] [I(), Z(0)]

>>> print(I.wires, Z.wires)
Wires([]) Wires([0])

Description of the Change:

The origin of the bug is that wire_order in PauliWord.operation was not being updated by the operators current wires set. This is strictly relevant to the Identity operator as it makes use of the argument.

Benefits: Identity.wires are now properly tracked when using Sum.terms.

import pennylane as qml

hamiltonian = qml.dot(
        [1, -0.5, 0.5], [qml.Identity(wires=[0, 1]), qml.PauliZ(0), qml.PauliZ(0)]
)
coeffs, ops = hamiltonian.terms()
I, Z = ops

>>> print(coeffs, ops)
[1.0, np.float64(0.0)] [I(), Z(0)]

>>> print(I.wires, Z.wires)
Wires([0,1]) Wires([0])

Possible Drawbacks: None

[sc-81355]

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.60%. Comparing base (37abd58) to head (c8d6838).
Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #6750   +/-   ##
=======================================
  Coverage   99.60%   99.60%           
=======================================
  Files         476      476           
  Lines       45210    45210           
=======================================
  Hits        45033    45033           
  Misses        177      177           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[JerryChen97]

Can we add a small test that makes use of the wire info of Identity? I feel that without an explicit example people would always wonder 'why do we need the wires of an Identity'.

[andrijapau]

Hmm, this might have some hidden side-effects that aren't being caught in the test suite. The gradient shape should be of length two here (one per coefficient), but instead I am getting four? 😅

import pennylane as qml

time, n, order = (4.2, 10, 4)

time = qml.numpy.array(time)
hamiltonian = qml.dot(
        [1, 1], [qml.I([0,1]), qml.PauliX(0)]
)

coeffs, ops = hamiltonian.terms()

@qml.qnode(qml.device("default.qubit"), diff_method="backprop")
def circ_bp(coeffs, time):
    with qml.queuing.QueuingManager.stop_recording():
        hamiltonian = qml.dot(coeffs, ops)

    qml.TrotterProduct(hamiltonian, time, n, order)
    return qml.probs()

>>> qml.jacobian(circ_bp)(coeffs, time)
[-0.85459891  0.          0.85459891  0.        ]
# [-0.85459891  0.85459891] on master branch

[mudit2812]

@andrijapau I actually think that the new behaviour is correct. The hamiltonian you created acts on 2 wires, so the output of qml.probs should have 4 elements.

[JerryChen97]

@andrijapau I actually think that the new behaviour is correct. The hamiltonian you created acts on 2 wires, so the output of qml.probs should have 4 elements.

So the wire info of Identity ops grant us the ability to extend the range of other local ops?

[andrijapau]

@andrijapau I actually think that the new behaviour is correct. The hamiltonian you created acts on 2 wires, so the output of qml.probs should have 4 elements.

So the wire info of Identity ops grant us the ability to extend the range of other local ops?

What do you mean by this?

[JerryChen97]

@andrijapau I actually think that the new behaviour is correct. The hamiltonian you created acts on 2 wires, so the output of qml.probs should have 4 elements.

So the wire info of Identity ops grant us the ability to extend the range of other local ops?

What do you mean by this?

Oh nm I'm just thinking that previous results are only showing the non-Identity operators' scope but the new ones reflect a larger scope extended by the Identity, which is actually logically correct here